[Paper Review] Linear-Quadratic Mean Field Social Optimization with a Major Player

Minyi Huang, Son Luu Nguyen|arXiv (Cornell University)|Apr 6, 2019

Mathematical Biology Tumor Growth43 references4 citations

TL;DR

This paper proposes a linear-quadratic mean field social optimization framework with a major player and a large number of minor players, where dynamics and costs depend on random parameters. By applying the person-by-person optimality principle and consistent mean field approximations, it derives a system of forward-backward stochastic differential equations (FBSDEs) that yield decentralized strategies achieving social optimality in large but finite populations.

ABSTRACT

This paper considers a linear-quadratic (LQ) mean field control problem involving a major player and a large number of minor players, where the dynamics and costs depend on random parameters. The objective is to optimize a social cost as a weighted sum of the individual costs under decentralized information. We apply the person-by-person optimality principle in team decision theory to the finite population model to construct two limiting variational problems whose solutions, subject to the requirement of consistent mean field approximations, yield a system of forward-backward stochastic differential equations (FBSDEs). We show the existence and uniqueness of a solution to the FBSDEs and obtain decentralized strategies nearly achieving social optimality in the original large but finite population model.

Motivation & Objective

• Address the gap in mean field control theory by incorporating a major player interacting with a large number of minor players, extending cooperative team decision models.
• Formulate a social optimization problem where the collective cost is minimized as a weighted sum of individual costs under decentralized information.
• Extend the person-by-person optimality principle from peer-level mean field teams to a mixed-player setting with random coefficients and coupled dynamics.
• Establish existence and uniqueness of solutions to the derived FBSDEs, ensuring decentralized strategies nearly achieve social optimality in finite populations.
• Generalize prior work by including random coefficients dependent on the major player's Brownian motion, modeling the major player as a common source of randomness.

Proposed method

• Model the system using linear-quadratic dynamics with random coefficients that depend on the major player's Brownian motion, introducing common noise.
• Apply the person-by-person optimality principle from team decision theory to derive two limiting variational problems for the major and minor players.
• Construct a system of forward-backward stochastic differential equations (FBSDEs) by enforcing consistency between the mean field approximation and the optimal control strategies.
• Use linear backward stochastic differential equation (BSDE) techniques to handle random coefficients and mean field terms in a unified framework.
• Derive explicit expressions for the optimal control strategies via state-space augmentation and solution of the FBSDEs, ensuring consistency with the mean field limit.
• Validate the solution through variational analysis, showing that perturbations in minor players' controls lead to consistent optimality conditions in the limit.

Experimental results

Research questions

• RQ1 How can social optimality be achieved in a mean field control system with a major player and a large number of minor players under random coefficients?
• RQ2 What is the role of the person-by-person optimality principle in extending team decision theory to mixed-player mean field models with random dynamics?
• RQ3 How do random coefficients dependent on the major player’s Brownian motion affect the structure of the optimal control strategies?
• RQ4 Can decentralized strategies be constructed that achieve social optimality in finite but large populations, and what is the asymptotic performance loss?
• RQ5 What is the mathematical structure of the FBSDE system that characterizes the optimal solution in this mixed-player social optimization framework?

Key findings

• The paper establishes the existence and uniqueness of a solution to the derived system of forward-backward stochastic differential equations (FBSDEs), ensuring a well-defined optimal control policy.
• The solution yields decentralized strategies for minor players that achieve social optimality in the finite population model, with optimality loss vanishing as the population size tends to infinity.
• The major player’s Brownian motion acts as a common source of randomness, and the model captures this through random coefficients in the dynamics and cost functions.
• The person-by-person optimality principle is successfully extended to the mixed-player setting, enabling the derivation of necessary conditions for social optimality under partial information.
• The FBSDE system is solved using linear BSDE techniques, allowing unified treatment of random mean field terms and coefficients, with explicit expressions for the optimal control and adjoint processes.
• The analysis confirms that the social optimum is a specific Pareto optimum, and the performance of the decentralized strategies approaches the global social optimum in the large-population limit.